lightposition for surface plots doesn't work

The parameter lightposition for surface plots expects an x, y, z vector. This vector is not explained in the documentation, does not make sense in any way, and playing with it only produces aggravation. It appears to be completely broken. This parameter should be defined in spherical coordinates and explained clearly in the documentation.

Not only does trying different values for x, y, z not make sense, I had a surface plot defined over the domains in x and y (my coordinates) of 0 to 6. I then realized that the domain should be -3 to 3; when I made this change, the lighting changed radically and I can't figure out how to get back the original lighting without reverting to the incorrect domains.

(1) lightposition needs to be fixed.

(2) lightposition needs to be in spherical coordinates.

(3) lightposition needs to be properly documented.

(4) lightposition needs to have at least an option which has a definition in "head" space so that the lighting doesn't change when merely the domain of the data is changed.

Start by locating the surface-plot lightposition entry point and its documentation; the issue names no files or tests. Reproduce the lighting change when the x and y domains move from 0–6 to -3–3, then define the desired coordinate behavior. Done means the parameter behaves consistently, supports the requested positioning semantics, and is clearly documented.